Unmanned helicopter control method based on fuzzy neural network expansion state observer

Abstract

The invention discloses an unmanned helicopter control method based on a fuzzy neural network extended state observer, and belongs to the technical field of transverse unmanned helicopter flight control, and the structure of a transverse unmanned helicopter flight controller comprises an FNNESO-RBSC controller, a control strategy and a transverse unmanned helicopter flight dynamics model; the FNNESO-RBSC controller calculates transverse, longitudinal, course and vertical channel control quantities according to system input and output information, the control quantities are distributed to all control surfaces of the transverse unmanned helicopter through a control strategy and act on a controlled object of the transverse unmanned helicopter, object state response of the transverse unmanned helicopter is changed under the action of aerodynamic force, and the control surfaces of the transverse unmanned helicopter are controlled. And the target flight state is controlled. According to the method, the fuzzy neural network is adopted to transform the extended state observer, and the adaptability of the extended state observer is improved, so that the unknown total disturbance needing to be compensated in the backstepping control law is more effectively obtained.

Description

Technical Field

[0001] The invention belongs to the technical field of flight control of horizontal unmanned helicopters, and in particular relates to an unmanned helicopter control method based on a fuzzy neural network extended state observer. Background Art

[0002] With the rapid development of microelectronics technology, unmanned helicopters are widely used in agricultural aviation operations such as sowing, remote sensing, and pesticide application. Compared with traditional symmetrical multi-rotor unmanned helicopters, horizontal unmanned helicopters have the advantages of wide spraying area and high wind-fog synchronization rate, so they have broader application prospects in the field of agricultural aviation. However, the horizontal unmanned helicopter is a complex nonlinear strong coupling system. The changes in weight, center of gravity and inertia during flight form internal disturbances of the uncertain system. At the same time, the horizontal unmanned helicopter will also be affected or threatened by external wind disturbances, which constitute external disturbances of the uncertain system. Complex uncertain internal and external disturbances seriously affect the stability of the attitude and position control of the horizontal unmanned helicopter, which is not conducive to safe operation flight.

[0003] Researchers have proposed many control methods for the stable flight of transverse unmanned helicopters. For example, linear control methods, including proportional integral derivative (PID) and linear quadratic regulator (LQR), are known for their simple design and easy engineering implementation. However, linear control faces challenges in effectively mitigating the disturbances encountered by transverse unmanned helicopters. In contrast, nonlinear control methods provide more effective solutions to the impact of disturbances. For example, backstepping control is known for its robustness, stability, and fast response design, and has been widely recognized in unmanned helicopter flight control. Liu et al. designed a backstepping fault-tolerant control method for the rotor tilt axis jamming fault of a transverse unmanned helicopter to ensure the stability of cruise flight, but this method requires accurate fault information to be determined. Although traditional backstepping control has good dynamic characteristics, its ability to overcome uncertain disturbances is not strong enough. To this end, Liang et al. designed an adaptive adjustment function to compensate for system uncertainty, designed a robust term function to solve the approximate error problem, and improved the anti-disturbance of backstepping control. This method has been effectively verified in the attitude control simulation of a transverse unmanned helicopter. Wang et al. proposed an adaptive command filtering backstepping sliding mode control method, which used an adaptive control strategy to estimate the upper limit of uncertainty and disturbance, thereby eliminating the influence of uncertain disturbance on the backstepping control accuracy.

[0004] Although backstepping control is more robust to system uncertainties and disturbances than traditional control methods, it generally requires partial unknown information to design the control law, which poses a challenge to practical applications. Therefore, alternative methods must be explored to reduce the dependence of backstepping control on the information of the horizontal unmanned helicopter model. Considering the characteristics of internal and external disturbances, Han Jingqing designed active disturbance rejection control (ADRC). The basic concept of ADRC includes collectively referring to internal and external disturbances as total disturbances and estimating their values ​​using an extended state observer (ESO). ESO can effectively estimate the total disturbance without using the model information of the controlled object. ESO can eliminate the dependence of backstepping control on the model and enable the backstepping controller to operate independently of the unmanned helicopter model. However, the adaptability of ESO needs to be enhanced, especially when the total disturbance exceeds a predetermined threshold, the estimation performance deteriorates due to inappropriate parameters. Summary of the invention

[0005] In view of the defects in the prior art, the present invention proposes a method of robust backstepping control (Fuzzy Neural Networks Extended State Observer-based Robust Backstepping Control, FNNESO-RBSC) based on fuzzy neural network extended state observer. The present invention integrates the advantages of neural network, fuzzy control, extended state observer and backstepping control, uses fuzzy neural network to transform the extended state observer, improves the adaptability of the extended state observer, and thus more effectively obtains the unknown total disturbance that needs to be compensated in the backstepping control law. FNNESO-RBSC can overcome the influence of internal and external disturbances of horizontal unmanned helicopters and meet the flight stability control requirements of horizontal unmanned helicopters.

[0006] The present invention is achieved in that:

[0007] A control method for an unmanned helicopter based on a fuzzy neural network extended state observer is characterized in that a flight controller structure of a lateral unmanned helicopter includes a FNNESO-RBSC controller, a manipulation strategy and a flight dynamics model of the lateral unmanned helicopter; the FNNESO-RBSC controller calculates lateral, longitudinal, heading and vertical channel control quantities according to system input and output information, distributes them to various control surfaces of the lateral unmanned helicopter through the manipulation strategy, acts on the controlled object of the lateral unmanned helicopter, and changes the object state response of the lateral unmanned helicopter under the action of aerodynamic force, so that it is controlled to a target flight state.

[0008] Furthermore, in the flight dynamics model, a horizontal unmanned helicopter is used, and its aerodynamic components include rotors, wings, fuselage and vertical tail. The left and right rotors have the same parameters except for the opposite rotation directions. The coordinate system used in the calculation of the flight dynamics model includes a ground coordinate system (0D X D Y D Z D ), body coordinate system (O B X B Y B Z B ), wind axis (O V X V Y V Z V ) and propeller shaft system (O S X S Y S Z S ).

[0009] Further, the rotor model is specifically:

[0010] The invention comprises three parts: rotor induced velocity model, blade flapping motion model and rotor aerodynamic calculation. The rotor induced velocity model adopts the first-order harmonic linear Pitt-Peters dynamic inflow model to calculate its rotor induced velocity; the first-order Fourier series is used to expand the induced velocity into a function of the rotor spanwise position and azimuth angle:

[0011]

[0012] In the formula, v0,v c ,v s are the time-averaged dimensionless inflow term of the rotor induced velocity, the first-order dimensionless longitudinal inflow term, and the first-order dimensionless lateral inflow term, respectively. is the dimensionless term of the blade spanwise position, ψ f is the blade azimuth;

[0013] According to the Pitt-Peters dynamic inflow model, the induced velocity component of the propeller shaft and propeller disk does not change with time under steady flight conditions, that is:

[0014]

[0015] In the formula, C T ,C L ,C M are the rotor lift coefficient, rolling moment coefficient and pitching moment coefficient, M and L respectively. NL is the gain matrix;

[0016] The blade flapping motion model is used to solve the rotor flapping angle component. Considering only the first-order rigid flapping of the blade, the first-order dynamic model of the flapping angle β is:

[0017] β=a0+a 1s cosψ f b 1s sinψ f(3)

[0018] In the formula, a0, a 1s ,b 1s are the rotor taper angle, rear chamfer angle and side chamfer angle respectively, and the expression is:

[0019]

[0020] Among them, Locke number γ is expressed as:

[0021]

[0022] In the formula, a ∞ is the slope of the lift line of the airfoil, b7 is the characteristic width of the blade, ρ kq is the air density, R is the rotor radius, λ0 is the rotor inflow ratio, μ is the rotor advance ratio, θ0 is the blade root installation angle, I β ,M β are the blade flapping hinge inertia moment and mass moment, A1 and B1 are the rotor collective pitch, lateral cyclic pitch and longitudinal cyclic pitch respectively;

[0023] Based on the induced velocity model and the flapping model, the forces and moments of the rotor in the body coordinate system are calculated according to the blade element theory.

[0024] Further, the wing model is specifically:

[0025] Considering the aerodynamic interference of the rotor wake on the wing, the wing is divided into two parts: the slipstream area and the freestream area. The slipstream area is the part affected by the rotor wake, and the freestream area is the part not affected by the rotor wake. The calculation process of the wing aerodynamic force is as follows: first, the slipstream area of ​​the wing is calculated according to the rotor advance ratio, and on this basis, the aerodynamic forces of the slipstream area and the freestream area under the wind axis are calculated respectively. The calculation method of the front wing and the rear wing is the same; finally, the forces and moments of the front and rear wings under the wind axis are projected to the body axis to obtain the total aerodynamic force of the wing;

[0026] Wing slipstream area S hl The calculation formula is as follows:

[0027]

[0028] Among them, the maximum wing slipstream area R max It is expressed as:

[0029]

[0030] In the formula, μ max is the advance ratio of the rotor wake just leaving the wing, μ is the rotor advance ratio, R is the rotor radius, C T is the rotor thrust coefficient, Z ris the dimensionless number of the vertical distance from the rotor rotation center to the wing;

[0031] Considering the influence of rotor induced speed, the airflow velocity at the aerodynamic center of the wing slipstream area is:

[0032]

[0033] Where V x ,V y ,V z ;ω x ,ω y ,ω z are the three-axis velocity and angular velocity of the center of mass of the tilt-rotating quadrotor UAV in the body axis system, x w ,y w ,z w are the coordinates of the wing aerodynamic center in the three directions of the body axis system, v i is the induced speed of the rotor at the aerodynamic center of the wing;

[0034] Dynamic pressure in the wing slipstream region q w,hl for:

[0035]

[0036] Lift in the wing slipstream region L w,hl , resistance D w,hl and pitching moment M wz,hl for:

[0037] L w,hl =q w,hl S hl C Lw,hl (12)

[0038] D w,hl =q w,hl S hl C Dw,hl (13)

[0039] M wz,hl =q w,hl S hl c w C Mw,hl (14)

[0040] In the formula, C Lw,hl ,C Dw,hl ,C Mw,hl are the lift coefficient, drag coefficient, and pitch moment coefficient of the wing under the wind axis system, S hl is the wing area in the slipstream region, c w is the chord length of the wing; then the force and moment of the wing slipstream region in the body coordinate system are obtained according to the coordinate projection;

[0041] The calculation method of aerodynamic force in the free flow area is the same as that in the slipstream area, except that the airflow velocity at the aerodynamic center of the wing free flow area does not include the rotor induced speed term.

[0042] Furthermore, the fuselage model is specifically:

[0043] The forces and moments of the fuselage are:

[0044]

[0045] Where D F ,L F ,S F They are the drag, lift and side force of the fuselage, M xF ,M yF ,M zF They are the rolling moment, yaw moment and pitch moment of the fuselage, l F is the characteristic length of the fuselage, A F is the characteristic area of ​​the fuselage, C DF ,C LF ,C SF They are the drag coefficient, lift coefficient and side force coefficient of the fuselage, C MxF ,C MyF ,C MzF They are the rolling moment coefficient, yaw moment coefficient and pitch moment coefficient of the fuselage respectively; then the force and moment of the fuselage in the fuselage coordinate system are obtained according to the coordinate projection.

[0046] Furthermore, the vertical tail model is specifically:

[0047] The tail power of the wind shaft system is:

[0048]

[0049] Where D V ,L V ,S V They are the drag, lift and side force of the vertical tail, C LV ,C DV are the lift coefficient and drag coefficient of the vertical tail, q v is the dynamic pressure of the vertical tail; then the force and moment of the vertical tail in the body coordinate system are obtained according to the coordinate projection.

[0050] Furthermore, the flight dynamics model of the horizontal unmanned helicopter is specifically as follows:

[0051] The aerodynamic resultant force and torque of the tilt-rotating quadrotor UAV under the body axis system can be obtained from the models of each aerodynamic component:

[0052]

[0053] In the formula, F x ,F y ,F z ,M x ,M y ,M z They are the combined external forces and combined external moments of the three axes of all components of the whole aircraft. According to the kinematic equation of the aircraft, the following nonlinear flight dynamics model of the horizontal unmanned helicopter is obtained:

[0054]

[0055] In the formula, ω x ,ω y ,ω z are the angular velocity in the body coordinate system, V x ,V y ,V z are the linear velocities in the body coordinate system, φ, θ, ψ are the roll, pitch and yaw Euler angles, X, Y, Z are the three-axis position coordinates in the ground coordinate system, I x ,I y ,I z is the moment of inertia, I xy is the product of inertia, g is the acceleration due to gravity;

[0056] Define X1=[φθψ], X2=[ω x ω y ω z ],X3=[V x V y V z ], X4=[XYZ], then equations (19) to (22) are simplified to:

[0057]

[0058] In the formula, F i (·)(i=1,2,3,4) represents the unknown function of the relevant state quantity, U a =[δ lat δ lon δ T ] represent the lateral, longitudinal and yaw channel control quantities of the rotor, respectively, U v =[δ φ δ θ δ col ] represent the virtual control amount of roll angle, pitch angle and vertical channel control amount respectively, B a ,B v Represent the unknown control gain matrices of attitude and velocity respectively.

[0059] Furthermore, the FNNESO-RBSC flight control law includes:

[0060] (1) Second-order nonlinear expansion system

[0061] Consider a second-order nonlinear uncertain object subject to external disturbances:

[0062]

[0063] Where u and y are the input and output signals of the system, f(x,w,t) is a nonlinear function containing unknown disturbances, and b(t) is an unknown compensation coefficient;

[0064] Let x3=f(x,w,t)+(b(t)-b0)u be the new expanded state variable, including the internal and external disturbances of the system, that is, the total disturbance, b0 be a constant, and record Then the system (24) becomes an extended system with three state variables, namely:

[0065]

[0066] (2) Fuzzy neural network extended state observer

[0067] The fuzzy neural network used to estimate the total disturbance is designed as a four-layer feedforward network containing a 2-dimensional input (x = [x1 x2]), n fuzzy rules and a 1-dimensional output (the total disturbance estimate);

[0068] First layer (input layer): pass the input variable x = [x1 x2] to the next layer;

[0069] The second layer (membership function layer): also known as the fuzzification layer, this layer receives the two outputs of the input layer. This paper uses the Gaussian function as the membership function to map the two variables x1 and x2 to n fuzzy rules respectively. The membership function value corresponding to the jth input variable under the i-th fuzzy rule is:

[0070]

[0071] In the formula, They represent the center and width of the membership function corresponding to the jth input variable under the i-th fuzzy rule respectively;

[0072] The third layer (fuzzy basis function layer): The fuzzy basis function corresponding to the i-th fuzzy rule is expressed as:

[0073]

[0074] The fourth layer (output layer): expressed as the product sum of fuzzy basis function and weight:

[0075]

[0076] In the formula, is the output of the fuzzy neural network, which represents the estimation of the total disturbance of the system. is the estimated weight matrix.

[0077] The total disturbance x3 is estimated by the fuzzy neural network, and the state quantities x1 and x2 are still estimated by ESO. Then the fuzzy neural network extended state observer (FNNESO) of the extended system (25) is:

[0078]

[0079] In the formula, λ, β i (i=1,2) is the gain coefficient, ζ is the fixed step size. Then the output variable of FNNESO can track the state variables and total disturbance of system (25), that is:

[0080]

[0081] (3) Robust backstepping control law

[0082] Define the error between the system target quantity and the actual quantity:

[0083] e1=x1-x g (32)

[0084]

[0085] In the formula, x g is the system target quantity;

[0086] 1) Define the Lyapunov function:

[0087]

[0088] definition Where c1 is a positive constant and z1 is a virtual control term, then:

[0089]

[0090] and

[0091]

[0092] Define the switching function as:

[0093] σ=k1e1+z1 (38)

[0094] Where k1>0;

[0095] because but:

[0096]

[0097] Since k1+c1>0, when σ=0, e1=0, z1=0 and To this end, the next step of design is needed;

[0098] 2) Define the Lyapunov function:

[0099]

[0100] but:

[0101]

[0102] The backstepping control law is designed as

[0103]

[0104] In the formula, h and c are positive constants;

[0105] Substituting the backstepping control law (42) into equation (41), we can obtain:

[0106]

[0107] Combined with the output of the fuzzy neural network extended state observer in equation (29), The estimation of the total disturbance x3 of the system can be obtained:

[0108]

[0109] Pick

[0110]

[0111] Let τ T =[e1 z1], then:

[0112]

[0113] If Q is a positive definite matrix, then:

[0114]

[0115] because:

[0116]

[0117] By adjusting the values ​​of h, c1, k1, |Q|>0, Q is guaranteed to be a positive definite matrix, thus ensuring

[0118] According to the LaSalle invariance principle, when taking When t→∞, e1≡0, z1≡0, σ≡0, then when t→∞, e1→0, z1→0, σ→0, and thus x1→x g,

[0119] In summary, the FNNESO-RBSC controller is obtained;

[0120] (4) FNNESO-RBSC control law

[0121] Combined with the nonlinear flight dynamics model of the horizontal unmanned helicopter represented by equation (23), its corresponding attitude, velocity and position models can be transformed into a second-order nonlinear uncertain object as shown in equation (24); therefore, the attitude, velocity and position control loops are designed.

[0122] Taking attitude control as an example, the control variable δ lat ,δ lon ,δ T It forms a single-input-single-output relationship with the state quantities φ, θ, ψ; the control quantity δ of the longitudinal channel lon Designed for:

[0123]

[0124] In the formula, θ g is the target pitch angle command signal, θ is the actual pitch angle of the horizontal unmanned helicopter, b 0θ is the control gain constant, is the estimated value of the pitch angle state and pitch angle rate, is the total disturbance estimate of the pitch angle channel; similarly, the roll angle control value δ can be obtained lat , heading angle control value δ T And the control quantity of speed and position control channels.

[0125] Furthermore, the manipulation strategy is specifically as follows:

[0126] The roll angle and lateral speed are adjusted by generating a rolling moment through the collective pitch difference of the left and right rotors; the pitch angle and longitudinal speed are adjusted by generating a pitch moment through the same longitudinal cyclic pitch control of the left and right rotors; the heading angle is adjusted by generating a yaw moment through the longitudinal cyclic pitch difference of the left and right rotors; the vertical speed is adjusted by changing the lift size through the same collective pitch control of the left and right rotors; thus, each control variable is shown in formula (50).

[0127]

[0128] In the formula, δ li ,δ ci (i=1,2) represent the longitudinal cyclic pitch change and collective pitch control values ​​of the two rotors respectively.

[0129] The beneficial effects of the present invention compared with the prior art are:

[0130] (1) Fuzzy neural networks can effectively improve the adaptability of the extended state observer and enhance the ability to estimate the unknown total disturbance faced by the unmanned helicopter.

[0131] (2) The FNNESO-RBSC controller has a strong ability to resist uncertain internal and external disturbances, and can quickly eliminate the impact of disturbances, ensure attitude stability, and ensure stable flight of the horizontal unmanned helicopter.

[0132] (3) Under the influence of strong internal and external disturbances, the FNNESO-RBSC controller can still ensure high-precision trajectory tracking control effect, and its anti-disturbance and robustness are better than those of traditional backstepping control and active disturbance rejection control. BRIEF DESCRIPTION OF THE DRAWINGS

[0133] Figure 1 The block diagram of the horizontal unmanned helicopter flight control system of the present invention;

[0134] Figure 2 The appearance of the horizontal unmanned helicopter of the present invention;

[0135] Figure 3 The horizontal unmanned helicopter coordinate system of the present invention;

[0136] Figure 4 Calculate the aerodynamic characteristics of the rotor of the present invention;

[0137] Figure 5 The present invention is a flow chart of wing aerodynamic calculation;

[0138] Figure 6 It is the structural diagram of the fuzzy neural network of the present invention;

[0139] Figure 7 The FNNESO-RBSC controller of the present invention;

[0140] Figure 8 is a block diagram of the attitude control loop of the present invention;

[0141] Fig. 9 It is the speed control loop block diagram of the present invention;

[0142] Fig.10 is a block diagram of a position control loop of the present invention;

[0143] Fig.11 It is a schematic diagram of the manipulation strategy of the present invention;

[0144] Fig.12 is the target trajectory curve of the present invention; wherein (a) is a three-dimensional trajectory curve, and (b) is a two-dimensional trajectory curve;

[0145] Fig.13The flight state response curves of the present invention are as follows; wherein (a) is the position response curve, (b) is the speed response curve, and (c) is the attitude response curve. DETAILED DESCRIPTION

[0146] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail by enumerating examples below. It should be noted that the specific implementation described here is only used to explain the present invention and is not used to limit the present invention.

[0147] The structure of the horizontal unmanned helicopter flight controller is as follows: Figure 1 As shown in the figure, it includes three parts: FNNESO-RBSC controller, control strategy and flight dynamics model of horizontal unmanned helicopter. FNNESO-RBSC controller calculates the lateral, longitudinal, heading and vertical channel control quantities according to the system input and output information, and distributes them to the control surfaces of the horizontal unmanned helicopter through the control strategy, acting on the controlled object of the horizontal unmanned helicopter, and changes the object state response of the horizontal unmanned helicopter under the action of aerodynamic force, so that it is controlled to the target flight state.

[0148] 1. Flight dynamics model

[0149] This paper studies the horizontal unmanned helicopter Figure 2 As shown, its aerodynamic components are mainly composed of rotors, wings, fuselage and vertical tail. Except for the opposite rotation directions of the left and right rotors, the other parameters are the same. Figure 3 The coordinate systems used in the model calculations are shown in D X D Y D Z D ), body coordinate system (O B X B Y B Z B ), wind axis system (O V X V Y V Z V ) and propeller shaft system (O S X S Y S Z S ). The main structural parameters of the horizontal unmanned helicopter are shown in Table 1.

[0150] Table 1 Main parameters of horizontal unmanned helicopter

[0151]

[0152] (1) Rotor model

[0153] The rotor model includes three parts: rotor induced velocity model, blade flapping motion model and rotor aerodynamic calculation. Figure 4 shown.

[0154] The rotor induced velocity model uses the first-order harmonic linear Pitt-Peters dynamic inflow model to calculate its rotor induced velocity. The induced velocity is expanded into a function of the rotor spanwise position and azimuth angle using the first-order Fourier series:

[0155]

[0156] In the formula, v0,v c ,v s are the time-averaged dimensionless inflow term of the rotor induced velocity, the first-order dimensionless longitudinal inflow term, and the first-order dimensionless lateral inflow term, respectively. is the dimensionless term of the blade spanwise position, ψ f is the blade azimuth.

[0157] According to the Pitt-Peters dynamic inflow model, the induced velocity component of the propeller shaft and propeller disk does not change with time under steady flight conditions, that is:

[0158]

[0159] In the formula, C T ,C L ,C M are the rotor lift coefficient, rolling moment coefficient and pitching moment coefficient, M and L respectively. NL is the gain matrix.

[0160] The blade flapping motion model is used to solve the rotor flapping angle component. Considering only the first-order rigid flapping of the blade, the first-order dynamic model of the flapping angle β is:

[0161] β=a0+a 1s cosψ f b 1s sinψ f (3)

[0162] In the formula, a0, a 1s ,b 1s are the rotor taper angle, rear chamfer angle and side chamfer angle respectively, and the expression is:

[0163]

[0164] Among them, Locke number γ is expressed as:

[0165]

[0166] In the formula, a ∞ is the slope of the lift line of the airfoil, b7 is the characteristic width of the blade, ρ kqis the air density, R is the rotor radius, λ0 is the rotor inflow ratio, μ is the rotor advance ratio, θ0 is the blade root installation angle, I β ,M β are the blade flapping hinge inertia moment and mass moment, A1 and B1 are the rotor collective pitch, lateral cyclic pitch and longitudinal cyclic pitch respectively.

[0167] Based on the induced velocity model and the flapping model, the forces and moments of the rotor in the body coordinate system are calculated according to the blade element theory.

[0168] (2) Wing model

[0169] Considering the aerodynamic interference of the rotor wake on the wing, the wing is divided into two parts: the slipstream area and the freeflow area. The slipstream area is the part affected by the rotor wake, and the freeflow area is the part not affected by the rotor wake. Figure 5 The figure shows the calculation process of wing aerodynamic force. First, the slipstream area of ​​the wing is calculated according to the rotor advance ratio. On this basis, the aerodynamic forces of the slipstream area and freestream area under the wind axis are calculated respectively. The calculation method of the front wing and the rear wing is the same. Finally, the forces and moments of the front and rear wings under the wind axis are projected to the body axis to obtain the total aerodynamic force of the wing.

[0170] Wing slipstream area S hl The calculation formula is as follows:

[0171]

[0172] Among them, the maximum wing slipstream area R max It is expressed as:

[0173]

[0174] In the formula, μ max is the advance ratio of the rotor wake just leaving the wing, μ is the rotor advance ratio, R is the rotor radius, C T is the rotor thrust coefficient, Z r It is a dimensionless number representing the vertical distance from the rotor's center of rotation to the wing.

[0175] Considering the influence of rotor induced speed, the airflow velocity at the aerodynamic center of the wing slipstream area is:

[0176]

[0177] Where V x ,V y ,V z ;ω x ,ω y ,ω z are the three-axis velocity and angular velocity of the center of mass of the tilt-rotating quadrotor UAV in the body axis system, xw ,y w ,z w are the coordinates of the wing aerodynamic center in the three directions of the body axis system, v i is the induced speed of the rotor at the aerodynamic center of the wing.

[0178] Dynamic pressure in the wing slipstream region q w,hl for:

[0179]

[0180] Lift in the wing slipstream region L w,hl , resistance D w,hl and pitching moment M wz,hl for:

[0181] L w,hl =q w,hl S hl C Lw,hl (12)

[0182] D w,hl =q w,hl S hl C Dw,hl (13)

[0183] M wz,hl =q w,hl S hl c w C Mw,hl (14)

[0184] In the formula, C Lw,hl ,C Dw,hl ,C Mw,hl are the lift coefficient, drag coefficient, and pitch moment coefficient of the wing under the wind axis system, S hl is the wing area in the slipstream region, c w is the chord length of the wing. Then, according to the coordinate projection, the forces and moments of the wing slipstream region in the body coordinate system are obtained.

[0185] The calculation method of aerodynamic force in the free flow area is the same as that in the slipstream area, except that the airflow velocity at the aerodynamic center of the wing free flow area does not include the rotor induced velocity term, which will not be repeated here.

[0186] (3) Fuselage model

[0187] The forces and moments of the fuselage are:

[0188]

[0189] Where D F ,L F ,S F They are the drag, lift and side force of the fuselage, MxF ,M yF ,M zF They are the rolling moment, yaw moment and pitch moment of the fuselage, l F is the characteristic length of the fuselage, A F is the characteristic area of ​​the fuselage, C DF ,C LF ,C SF They are the drag coefficient, lift coefficient and side force coefficient of the fuselage, C MxF ,C MyF ,C MzF They are the rolling moment coefficient, yaw moment coefficient and pitch moment coefficient of the fuselage. Then, the force and moment of the fuselage in the fuselage coordinate system are obtained according to the coordinate projection.

[0190] (4) Vertical tail model

[0191] The tail power of the wind shaft system is:

[0192]

[0193] Where D V ,L V ,S V They are the drag, lift and side force of the vertical tail, C LV ,C DV are the lift coefficient and drag coefficient of the vertical tail, q v is the dynamic pressure of the vertical tail. Then, according to the coordinate projection, we can get the force and moment of the vertical tail in the body coordinate system.

[0194] (5) Nonlinear flight dynamics model of the whole aircraft

[0195] The aerodynamic resultant force and torque of the tilt-rotating quadrotor UAV under the body axis system can be obtained from the aerodynamic models of the above components:

[0196]

[0197] In the formula, F x ,F y ,F z ,M x ,M y ,M z They are the combined external forces and combined external moments of the three axes of all components of the whole aircraft. According to the kinematic equation of the aircraft, the following nonlinear flight dynamics model of the horizontal unmanned helicopter is obtained:

[0198]

[0199]

[0200] In the formula, ω x ,ω y,ω z are the angular velocity in the body coordinate system, V x ,V y ,V z are the linear velocities in the body coordinate system, φ, θ, ψ are the roll, pitch and yaw Euler angles, X, Y, Z are the three-axis position coordinates in the ground coordinate system, I x ,I y ,I z is the moment of inertia, I xy is the product of inertia, and g is the acceleration due to gravity.

[0201] Define X1=[φθψ], X2=[ω x ω y ω z ],X3=[V x V y V z ], X4=[XYZ], then equations (19) to (22) are simplified to:

[0202]

[0203] In the formula, F i (·)(i=1,2,3,4) represents the unknown function of the relevant state quantity, U a =[δ lat δ lon δ T ] represent the lateral, longitudinal and yaw channel control quantities of the rotor, respectively, U v =[δ φ δ θ δ col ] represent the virtual control amount of roll angle, pitch angle and vertical channel control amount respectively, B a ,B v Represent the unknown control gain matrices of attitude and velocity respectively.

[0204] 2. FNNESO-RBSC flight control law

[0205] (1) Second-order nonlinear expansion system

[0206] Consider a second-order nonlinear uncertain object subject to external disturbances:

[0207]

[0208] Where u and y are the system input and output signals, f(x,w,t) is a nonlinear function containing unknown disturbances, and b(t) is an unknown compensation coefficient.

[0209] Let x3=f(x,w,t)+(b(t)-b0)u be the new expanded state variable, including the internal and external disturbances of the system, that is, the total disturbance, b0 be a constant, and record Then the system (24) becomes an extended system with three state variables, namely:

[0210]

[0211] (2) Fuzzy Neural Network Extended State Observer

[0212] The fuzzy neural network used to estimate the total disturbance is designed as a four-layer feedforward network, such as Figure 6 As shown, it contains 2-dimensional input (x=[x1 x2]), n fuzzy rules and one-dimensional output (total disturbance estimate).

[0213] First layer (input layer): Pass the input variable x = [x1 x2] to the next layer.

[0214] The second layer (membership function layer): also known as the fuzzification layer, this layer receives the two outputs of the input layer. This paper uses the Gaussian function as the membership function to map the two variables x1 and x2 to n fuzzy rules respectively. The membership function value corresponding to the jth input variable under the i-th fuzzy rule is:

[0215]

[0216] In the formula, They respectively represent the center and width of the membership function corresponding to the j-th input variable under the i-th fuzzy rule.

[0217] The third layer (fuzzy basis function layer): The fuzzy basis function corresponding to the i-th fuzzy rule is expressed as:

[0218]

[0219] The fourth layer (output layer): expressed as the product sum of fuzzy basis function and weight:

[0220]

[0221] In the formula, is the output of the fuzzy neural network, which represents the estimation of the total disturbance of the system. is the estimated weight matrix.

[0222] The total disturbance x3 is estimated by the fuzzy neural network, and the state quantities x1 and x2 are still estimated by ESO. Then the fuzzy neural network extended state observer (FNNESO) of the extended system (25) is:

[0223]

[0224] In the formula, λ, β i (i=1,2) is the gain coefficient, ζ is the fixed step size. Then the output variable of FNNESO can track the state variables and total disturbance of system (25), that is:

[0225]

[0226] (3) Robust backstepping control law

[0227] Define the error between the system target quantity and the actual quantity:

[0228] e1=x1-x g (32)

[0229]

[0230] In the formula, x g is the system target quantity.

[0231] 1) Define the Lyapunov function:

[0232]

[0233] definition Where c1 is a positive constant and z1 is a virtual control term, then:

[0234]

[0235] and

[0236]

[0237] Define the switching function as:

[0238] σ=k1e1+z1 (38)

[0239] In the formula, k1>0.

[0240] because but:

[0241]

[0242] Since k1+c1>0, when σ=0, e1=0, z1=0 and To this end, the next step of design is needed.

[0243] 2) Define the Lyapunov function:

[0244]

[0245] but:

[0246]

[0247] The backstepping control law is designed as

[0248]

[0249] Where h and c are positive constants.

[0250] Substituting the backstepping control law (42) into equation (41), we can obtain:

[0251]

[0252] Combined with the output of the fuzzy neural network extended state observer in equation (29), The estimation of the total disturbance x3 of the system can be obtained:

[0253]

[0254] Pick

[0255]

[0256] Let τ T =[e1 z1], then:

[0257]

[0258] If Q is a positive definite matrix, then:

[0259]

[0260] because:

[0261]

[0262] By adjusting the values ​​of h, c1, k1, Q>0, thus ensuring that Q is a positive definite matrix, thus ensuring

[0263] According to the LaSalle invariance principle, when taking When t→∞, e1≡0, z1≡0, σ≡0, then when t→∞, e1→0, z1→0, σ→0, and thus x1→x g ,

[0264] In summary, FNNESO-RBSC controller is Figure 7 shown.

[0265] (4) FNNESO-RBSC control law

[0266] Combined with the nonlinear flight dynamics model of the horizontal unmanned helicopter represented by equation (23), its corresponding attitude, velocity and position models can be transformed into a second-order nonlinear uncertain object as shown in equation (24). Figures 8 to 10 The attitude, velocity, and position control loops are shown.

[0267] Taking attitude control as an example, the control variable δ lat ,δ lon ,δ T It forms a single-input-single-output relationship with the state quantities φ, θ, ψ. The control quantity δ of the longitudinal channel lon Designed for:

[0268]

[0269] In the formula, θ g is the target pitch angle command signal, θ is the actual pitch angle of the horizontal unmanned helicopter, b 0θ is the control gain constant, is the estimated value of the pitch angle state and pitch angle rate, is the estimated value of the total disturbance of the pitch angle channel.

[0270] Similarly, the roll angle control value δ can be obtained lat , heading angle control value δ T As well as the control quantities of the speed and position control channels, they will not be elaborated on here.

[0271] 3. Manipulation strategy

[0272] The control strategy of horizontal unmanned helicopter is as follows Fig.11 As shown in , the roll moment is generated by the differential pitch of the left and right rotors to adjust the roll angle and lateral speed; the pitch moment is generated by the same longitudinal cyclic pitch control of the left and right rotors to adjust the pitch angle and longitudinal speed; the yaw moment is generated by the longitudinal cyclic pitch differential of the left and right rotors to adjust the heading angle; the vertical speed is adjusted by changing the lift size through the same collective pitch control of the left and right rotors. Therefore, each control variable is shown in formula (50).

[0273]

[0274] In the formula, δ li ,δ ci (i=1,2) represent the longitudinal cyclic pitch and collective pitch control values ​​of the two rotors respectively.

[0275] 4. Simulation verification experiment

[0276] In order to verify the effectiveness of the FNNESO-RBSC controller designed in this paper, attitude control simulation experiments are carried out in the MATLAB / Simulink environment. Fig.12As shown in the figure, the target trajectory includes two parts: the figure-8 maneuver and the ski maneuver. In the simulation test, the controllers used are the BSC controller, the ADRC controller and the FNNESO-RBSC controller designed by the present invention. A rectangular wind with a wind speed of 9m / s and a pulse width of 1 second is added to the three-axis flow direction of the rotor as an external disturbance, and the total weight and the three-axis inertia product of the horizontal unmanned helicopter are uniformly reduced by 15% over time within the initial 30 seconds. The simulation results are shown in Fig.13 As shown. From the position tracking curve, it can be seen that the three controllers can track the target flight path well, but the position error curve shows that the FNNESO-RBSC controller has the smallest tracking error and the highest tracking accuracy. At the same time, due to the influence of external wind disturbance and internal disturbance, the speed and attitude response curves of the three controllers have a certain degree of oscillation and fluctuation, but the flight state of the FNNESO-RBSC controller is least affected by disturbances and the flight is the most stable. The above analysis shows that the trajectory tracking control performance of the FNNESO-RBSC controller is better, it can effectively overcome the influence of internal and external disturbances, and it has strong robustness and anti-disturbance.

[0277] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0278] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be regarded as within the protection scope of the present invention.

Claims

1. A control method for an unmanned helicopter based on a fuzzy neural network extended state observer, characterized in that: The flight controller structure of the horizontal unmanned helicopter includes the FNNESO-RBSC controller, the control strategy and the flight dynamics model of the horizontal unmanned helicopter; The FNNESO-RBSC controller calculates the lateral, longitudinal, heading and vertical channel control quantities according to the system input and output information, distributes them to the control surfaces of the horizontal unmanned helicopter through the control strategy, acts on the controlled object of the horizontal unmanned helicopter, and changes the object state response of the horizontal unmanned helicopter under the action of aerodynamic force, so that it is controlled to the target flight state.

2. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 1 is characterized in that: In the flight dynamics model, a horizontal unmanned helicopter is used, and its aerodynamic components include rotors, wings, fuselage and vertical tail; the coordinate system used in the calculation of the flight dynamics model includes a ground coordinate system (0 D X D Y D Z D ), body coordinate system (O B X B Y B Z B ), wind axis system (O V X V Y V Z V ) and propeller shaft system (O S X S Y S Z S ).

3. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 2 is characterized in that: The rotor model is specifically: The invention comprises three parts: rotor induced velocity model, blade flapping motion model and rotor aerodynamic calculation. The rotor induced velocity model adopts the first-order harmonic linear Pitt-Peters dynamic inflow model to calculate its rotor induced velocity; the first-order Fourier series is used to expand the induced velocity into a function of the rotor spanwise position and azimuth angle: In the formula, v0,v c ,v s are the time-averaged dimensionless inflow term of the rotor induced velocity, the first-order dimensionless longitudinal inflow term, and the first-order dimensionless lateral inflow term, respectively. is the dimensionless term of the blade spanwise position, ψ f is the blade azimuth; According to the Pitt-Peters dynamic inflow model, the induced velocity component of the propeller shaft and propeller disk does not change with time under steady flight conditions, that is: In the formula, C T ,C L ,C M are the rotor lift coefficient, rolling moment coefficient and pitching moment coefficient, M and L respectively. NL is the gain matrix; The blade flapping motion model is used to solve the rotor flapping angle component. Considering only the first-order rigid flapping of the blade, the first-order dynamic model of the flapping angle β is: β=a0+a 1s cosψ f b 1s sinψ f (3) In the formula, a0, a 1s ,b 1s are the rotor taper angle, rear chamfer angle and side chamfer angle respectively, and the expression is: Among them, Locke number γ is expressed as: In the formula, a ∞ is the slope of the lift line of the airfoil, b7 is the characteristic width of the blade, ρ kq is the air density, R is the rotor radius, λ0 is the rotor inflow ratio, μ is the rotor advance ratio, θ0 is the blade root installation angle, I β ,M β are the blade flapping hinge inertia moment and mass moment, A1 and B1 are the rotor collective pitch, lateral cyclic pitch and longitudinal cyclic pitch respectively; Based on the induced velocity model and the flapping model, the forces and moments of the rotor in the body coordinate system are calculated according to the blade element theory.

4. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 2 is characterized in that: The wing model is specifically: Considering the aerodynamic interference of the rotor wake on the wing, the wing is divided into two parts: the slipstream area and the freestream area. The slipstream area is the part affected by the rotor wake, and the freestream area is the part not affected by the rotor wake. The calculation process of the wing aerodynamic force is as follows: first, the slipstream area of ​​the wing is calculated according to the rotor advance ratio, and on this basis, the aerodynamic forces of the slipstream area and the freestream area under the wind axis are calculated respectively. The calculation method of the front wing and the rear wing is the same; finally, the forces and moments of the front and rear wings under the wind axis are projected to the body axis to obtain the total aerodynamic force of the wing; Wing slipstream area S hl The calculation formula is as follows: Among them, the maximum wing slipstream area R max It is expressed as: In the formula, μ max is the advance ratio of the rotor wake just leaving the wing, μ is the rotor advance ratio, R is the rotor radius, C T is the rotor thrust coefficient, Z r is the dimensionless number of the vertical distance from the rotor rotation center to the wing; Considering the influence of rotor induced speed, the airflow velocity at the aerodynamic center of the wing slipstream area is: Where V x ,V y ,V z ;ω x ,ω y ,ω z are the three-axis velocity and angular velocity of the center of mass of the tilt-rotating quadrotor UAV in the body axis system, x w ,y w ,z w are the coordinates of the wing aerodynamic center in the three directions of the body axis system, v i is the induced speed of the rotor at the aerodynamic center of the wing; Dynamic pressure in the wing slipstream region q w,hl for: Lift in the wing slipstream region L w,hl , resistance D w,hl and pitching moment M wz,hl for: L w,hl =q w,hl S hl C Lw,hl (12) D w,hl =q w,hl S hl C Dw,hl (13) M wz,hl =q w,hl S hl c w C Mw,hl (14) In the formula, C Lw,hl ,C Dw,hl ,C Mw,hl are the lift coefficient, drag coefficient, and pitch moment coefficient of the wing under the wind axis system, S hl is the wing area in the slipstream region, c w is the chord length of the wing; then the force and moment of the wing slipstream region in the body coordinate system are obtained according to the coordinate projection; The calculation method of aerodynamic force in the free flow area is the same as that in the slipstream area, except that the airflow velocity at the aerodynamic center of the wing free flow area does not include the rotor induced speed term.

5. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 2 is characterized in that: The fuselage model is specifically: The forces and moments of the fuselage are: Where D F ,L F ,S F They are the drag, lift and side force of the fuselage, M xF ,M yF ,M zF They are the rolling moment, yaw moment and pitch moment of the fuselage, l F is the characteristic length of the fuselage, A F is the characteristic area of ​​the fuselage, C DF ,C LF ,C SF They are the drag coefficient, lift coefficient and side force coefficient of the fuselage, C MxF ,C MyF ,C MzF They are the rolling moment coefficient, yaw moment coefficient and pitch moment coefficient of the fuselage respectively; then the force and moment of the fuselage in the fuselage coordinate system are obtained according to the coordinate projection.

6. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 2 is characterized in that: The vertical tail model is specifically: The tail power of the wind shaft system is: Where D V ,L V ,S V They are the drag, lift and side force of the vertical tail, C LV ,C DV are the lift coefficient and drag coefficient of the vertical tail, q v is the dynamic pressure of the vertical tail; then the force and moment of the vertical tail in the body coordinate system are obtained according to the coordinate projection.

7. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 1 is characterized in that: The flight dynamics model of the horizontal unmanned helicopter is specifically as follows: The aerodynamic resultant force and torque of the tilt-rotating quad-rotor unmanned aerial vehicle in the body axis system can be obtained from the models of each aerodynamic component: In the formula, F x ,F y ,F z ,M x ,M y ,M z They are the combined external forces and combined external moments of the three axes of all components of the whole aircraft. According to the kinematic equation of the aircraft, the following nonlinear flight dynamics model of the horizontal unmanned helicopter is obtained: In the formula, ω x ,ω y ,ω z are the angular velocity in the body coordinate system, V x ,V y ,V z are the linear velocities in the body coordinate system, φ, θ, ψ are the roll, pitch and yaw Euler angles, X, Y, Z are the three-axis position coordinates in the ground coordinate system, I x ,I y ,I z is the moment of inertia, I xy is the product of inertia, g is the acceleration due to gravity; Define X1=[φθψ], X2=[ω x ω y ω z ],X3=[V x V y V z ], X4=[XYZ], then equations (19) to (22) are simplified to: In the formula, F i (·)(i=1,2,3,4) represents the unknown function of the relevant state quantity, U a =[δ lat δ lon δ T ] represent the lateral, longitudinal and yaw channel control quantities of the rotor, respectively, U v =[δ φ δ θ δ col ] represent the virtual control amount of roll angle, pitch angle and vertical channel control amount respectively, B a ,B v Represent the unknown control gain matrices of attitude and velocity respectively.

8. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 1 is characterized in that: The FNNESO-RBSC flight control law includes: (1) Second-order nonlinear expansion system Consider a second-order nonlinear uncertain object subject to external disturbances: Where u and y are the input and output signals of the system, f(x,w,t) is a nonlinear function containing unknown disturbances, and b(t) is an unknown compensation coefficient; Let x3=f(x,w,t)+(b(t)-b0)u be the new expanded state variable, including the internal and external disturbances of the system, that is, the total disturbance, b0 be a constant, and record Then the system (24) becomes an extended system with three state variables, namely: (2) Fuzzy neural network extended state observer The fuzzy neural network used to estimate the total disturbance is designed as a four-layer feedforward network containing a 2-dimensional input (x = [x1 x2]), n fuzzy rules and a 1-dimensional output (the total disturbance estimate); First layer (input layer): pass the input variable x = [x1 x2] to the next layer; The second layer (membership function layer): also known as the fuzzification layer, this layer receives the two outputs of the input layer. This paper uses the Gaussian function as the membership function to map the two variables x1 and x2 to n fuzzy rules respectively. The membership function value corresponding to the jth input variable under the i-th fuzzy rule is: In the formula, They represent the center and width of the membership function corresponding to the jth input variable under the i-th fuzzy rule respectively; The third layer (fuzzy basis function layer): The fuzzy basis function corresponding to the i-th fuzzy rule is expressed as: The fourth layer (output layer): expressed as the product sum of fuzzy basis function and weight: In the formula, is the output of the fuzzy neural network, which represents the estimation of the total disturbance of the system. is the estimated weight matrix. The total disturbance x3 is estimated by the fuzzy neural network, and the state quantities x1 and x2 are still estimated by ESO. Then the fuzzy neural network extended state observer (FNNESO) of the extended system (25) is: In the formula, λ, β i (i=1,2) is the gain coefficient, ζ is the fixed step size. Then the output variable of FNNESO can track the state variables and total disturbance of system (25), that is: (3) Robust backstepping control law Define the error between the system target quantity and the actual quantity: e1=x1-x g (32) In the formula, x g is the system target quantity; 1) Define the Lyapunov function: definition Where c1 is a positive constant and z1 is a virtual control term, then: and Define the switching function as: σ=k1e1+z1 (38) Where, k1>0; because but: Since k1+c1>0, when σ=0, e1=0, z1=0 and To this end, the next step of design is needed; 2) Define the Lyapunov function: but: The backstepping control law is designed as In the formula, h, χ are positive constants; Substituting the backstepping control law (42) into equation (41), we can obtain: Combined with the output of the fuzzy neural network extended state observer in equation (29), The estimation of the total disturbance x3 of the system can be obtained: Pick Let τ T =[e1 z1], then: If Q is a positive definite matrix, then: because: By adjusting the values ​​of h, c1, k1, Q|>0, Q is guaranteed to be a positive definite matrix, thus ensuring According to the LaSalle invariance principle, when taking When t→∞, e1≡0, z1≡0, σ≡0, then when t→∞, e1→0, z1→0, σ→0, and thus x1→x g, In summary, the FNNESO-RBSC controller is obtained; (4) FNNESO-RBSC control law Combined with the nonlinear flight dynamics model of the horizontal unmanned helicopter represented by equation (23), its corresponding attitude, velocity and position models can be transformed into a second-order nonlinear uncertain object as shown in equation (24); therefore, the attitude, velocity and position control loops are designed. Taking attitude control as an example, the control variable δ lat ,δ lon ,δ T It forms a single-input-single-output relationship with the state quantities φ, θ, ψ; the control quantity δ of the longitudinal channel lon Designed for: In the formula, θ g is the target pitch angle command signal, θ is the actual pitch angle of the horizontal unmanned helicopter, b 0θ is the control gain constant, is the estimated value of the pitch angle state and pitch angle rate, is the total disturbance estimate of the pitch angle channel; similarly, the roll angle control value δ can be obtained lat , heading angle control value δ T And the control quantity of speed and position control channels.

9. The unmanned helicopter control method based on fuzzy neural network extended state observer according to claim 1 is characterized in that: The manipulation strategy is specifically as follows: The roll angle and lateral speed are adjusted by generating a rolling moment through the collective pitch difference of the left and right rotors; the pitch angle and longitudinal speed are adjusted by generating a pitch moment through the same longitudinal cyclic pitch control of the left and right rotors; the heading angle is adjusted by generating a yaw moment through the longitudinal cyclic pitch difference of the left and right rotors; the vertical speed is adjusted by changing the lift size through the same collective pitch control of the left and right rotors; thus, each control variable is as shown in formula (50). In the formula, δ li ,δ ci (i=1,2) represent the longitudinal cyclic pitch change and collective pitch control values ​​of the two rotors respectively.

Images